Fiftycharacters.One job.
Every ChatGPT ad is the same six-component card: name, favicon, a 50-character title, a 100-character description, a URL, and one square image. It sits below a conversation, so it has to read as the answer to that conversation. We write the copy, produce the images, and author the context hints that decide which conversation it appears under.
- Title limit
- 50
- Characters, per OpenAI's Ads Manager
- Description limit
- 100
- Characters — one proof point
- Image
- 1:1
- PNG or JPG, at least 256 × 256 px
- Creative cadence
- Weekly
- 3–6 new ads per drop on every retainer
What a ChatGPTad is made of.
OpenAI documents six building blocks and no more. Two of them are set once at account level and render in every ad, which is why we treat the advertiser name and logo as creative rather than admin.
| Component | Constraint | What we optimise |
|---|---|---|
| Advertiser name | Set in Settings → Account info | Spelled exactly as buyers should read it |
| Favicon (logo) | 1:1 image from Account info | High-contrast mark that survives 16px |
| Title | 50 characters maximum | One job: the outcome or comparison the user is mid-research on |
| Description | 100 characters maximum | The single most credible proof point |
| Landing page URL | One per ad, UTMs persist | Page that answers the same question the conversation did |
| Image | PNG or JPG, square, ≥ 256 × 256 px | Product or outcome shot; lifestyle 16:9 crops fail here |
Context hints,not keywords.
There is no keyword list in ChatGPT Ads. Each ad group carries context hints — short descriptions of the audience, intent, and topic of the conversations it should appear under. OpenAI matches ads to conversations by relevance, using the hints together with the title, the description, and the landing page. Creative and targeting are therefore one job, which is why we do both.
| Pattern | Matches | Example hint |
|---|---|---|
| Problem-first | Users describing a pain before they know the category | "Teams struggling to forecast cash flow across multiple entities" |
| Comparison | Users weighing named alternatives | "Comparing project-management tools for agencies of 20–100 people" |
| Job-to-be-done | Users asking how to accomplish a task | "Homeowners planning a kitchen remodel and asking about cabinet options" |
| Buyer-stage | Users signalling readiness to purchase | "Shoppers asking which magnesium supplement to buy for sleep" |
| Exclusion | Conversations you do not want | "Not students, not free-tier users, not job seekers" |
The creativesystem.
- 1
Map the conversations, not the keywords
We start from the questions your buyers actually ask an assistant, grouped by buyer moment. Each group becomes an ad group with its own context hints, written in the Audience–Intent–Topic anatomy OpenAI's system uses to match ads to conversations.
- 2
Write titles that answer the question
Fifty characters leaves room for one idea. We write titles as the direct answer to the conversation the ad sits under — the comparison, the outcome, or the specific product — and test three to six per ad group rather than one hero line.
- 3
Give the description one proof point
A hundred characters carries a single credible claim: a certification, a number, a guarantee, a delivery promise. Taglines, discounts, and proof points cannot all fit, so the description earns the click with the most defensible one.
- 4
Produce square, high-contrast images
The image is a small square next to text. We shoot or crop for legibility at thumbnail size — product on neutral background, or a single outcome — and supply variants so the image can be tested independently of copy.
- 5
Match the landing page to the conversation
OpenAI's relevance inputs include the landing page, not only the hints and copy. A generic homepage weakens matching and conversion at once. We pair each ad group with a page that continues the specific conversation.
- 6
Ship weekly, retire by cohort
New titles, descriptions, and images drop weekly. Winners and losers are judged on weekly cohorts with conversion data, never on day-one CTR, and every change is dated in the account change log.
What goodlooks like.
Three illustrative units, each written as the answer to the conversation it would sit under. Fictional brands; real constraints.
User asked: best magnesium for sleep?
Magnesium Glycinate Built for Deeper Sleep
Third-party tested. 400mg per serving. Free shipping over $40.
42/50 · 62/100
User asked: how to replace a spreadsheet sales pipeline
Pipeline CRM Teams Switch to From Sheets
Import your sheet in 10 minutes. No per-seat pricing under 25 users.
40/50 · 68/100
User asked: cost to replace kitchen cabinets
Kitchen Cabinet Quotes in 48 Hours
Licensed installers. Fixed-price quotes. 4.9★ from 2,100 remodels.
34/50 · 66/100
Answered
straight.
What does a ChatGPT ad look like?
A single compact, labeled card that appears below a relevant ChatGPT conversation, visually separate from the answer. It is built from six components: advertiser name, favicon, a title of up to 50 characters, a description of up to 100 characters, a landing page URL, and one square image. There is no second headline, sitelink, or callout extension.
Who sees ChatGPT ads?
Free and Go users in the markets where Ads Manager is live, including the United States, Canada, Australia, and New Zealand. Plus, Pro, and Business-plan users do not see ads, and neither do users OpenAI predicts to be under 18.
What are context hints in ChatGPT Ads?
Context hints are the targeting layer. Instead of keywords, you describe the audience, intent, and topic of the conversations an ad group should appear under, and OpenAI's relevance-weighted system matches ads to conversations using the hints, the ad title and copy, and the landing page.
How is ChatGPT ad copy different from Google Ads copy?
Shorter, and answering a conversation rather than a query. A Google responsive search ad has up to 15 headlines and 4 descriptions; a ChatGPT ad has one 50-character title and one 100-character description. The copy has to read as a direct response to what the user was just discussing, not as a slogan.
Can you use my existing Meta or Google creative?
Rarely as-is. Landscape and vertical video, 16:9 lifestyle images, and long headline sets do not fit the square, text-led unit. We adapt the proof points and offers, then rebuild the assets for this placement.
How many ads and context hints should each ad group have?
Several ads per ad group so titles and images can be tested against each other, and a tight set of context hints that describe one buyer moment. One ad in one ad group produces a result you cannot learn from.
Do you handle the advertiser name and logo too?
Yes. Both are set in Settings → Account info and render in every ad unit. We check spelling, casing, and logo contrast at favicon size before the first campaign goes live, because both are part of the creative whether or not anyone treats them that way.